Proliferative g enes dominate malignancy-risk gene ...

Proliferative genes dominatemalignancy-risk gene signature inhistologically-normal breast tissue

The Harvard community has made thisarticle openly available. Please share howthis access benefits you. Your story matters

Citation Chen, Dung-Tsa, Aejaz Nasir, Aedin Culhane, ChinnamballyVenkataramu, William Fulp, Renee Rubio, Tao Wang, et al.2009. “Proliferative Genes Dominate Malignancy-Risk GeneSignature in Histologically-Normal Breast Tissue.” Breast CancerResearch and Treatment 119 (2) (March 6): 335–346. doi:10.1007/s10549-009-0344-y.

Published Version doi:10.1007/s10549-009-0344-y

Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:29004170

Terms of Use This article was downloaded from Harvard University’s DASHrepository, and is made available under the terms and conditionsapplicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAA

http://osc.hul.harvard.edu/dash/open-access-feedback?handle=&title=Proliferative%20genes%20dominate%20malignancy-risk%20gene%20signature%20in%20histologically-normal%20breast%20tissue&community=1/4454687&collection=1/4454688&owningCollection1/4454688&harvardAuthors=ca70fb7de622a9e968d73b838d3ec8ef&department

http://nrs.harvard.edu/urn-3:HUL.InstRepos:29004170

http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAA



Proliferative genes dominate malignancy-risk gene signature inhistologically-normal breast tissue

Dung-Tsa Chen1, Aejaz Nasir*,2, Aedin Culhane3, Chinnambally Venkataramu5, WilliamFulp1, Renee Rubio3, Tao Wang4, Deepak Agrawal5, Susan M McCarthy5, Mike Gruidl5,Gregory Bloom1, Tove Anderson3, Joe White3, John Quackenbush3, and TimothyYeatman61Biostatistics Division, Moffitt Cancer Center & Research Institute, Tampa, FL 33612, USA.2Pathology, Moffitt Cancer Center & Research Institute, Tampa, FL 33612, USA.3 Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute Boston, MA02115, USA.4Department of Epidemiology and Biostatistics, University of South Florida, Tampa, FL 33612, USA.5Molecular Oncology, Moffitt Cancer Center & Research Institute, Tampa, FL 33612, USA.6Surgery and Interdisciplinary Oncology, Moffitt Cancer Center & Research Institute, Tampa, FL33612, USA.

AbstractPURPOSE—Historical data have indicated the potential for the histologically-normal breast toharbor pre-malignant changes at the molecular level. We postulated that a histologically-normaltissue with “tumor-like” gene expression pattern might harbor substantial risk for future cancerdevelopment. Genes associated with these high-risk tissues were considered to be “malignancy-riskgenes”.

EXPERIMENTAL DESIGN—From a total of 90 breast cancer patients, we collected a set of 143histologically-normal breast tissues derived from patients harboring breast cancer who underwentcurative mastectomy, as well as a set of 42 invasive ductal carcinomas (IDC) of various histologicgrades. All samples were assessed for global gene expression differences using microarray analysis.For the purpose of this study we defined normal breast tissue to include histologically normal andbenign lesions.

RESULTS—Here we report the discovery of a “malignancy-risk” gene signature that may portendrisk of breast cancer development in benign, but molecularly-abnormal, breast tissue. Pathwayanalysis showed that the malignancy-risk signature had a dramatic enrichment for genes withproliferative function, but appears to be independent of ER, PR, and HER2 status. The signature wasvalidated by RT-PCR, with a high correlation (Pearson correlation=0.95 with p<0.0001) withmicroarray data.

CONCLUSION—These results suggest a predictive role for the malignancy-risk signature in normalbreast tissue. Proliferative biology dominates the earliest stages of tumor development.

While breast cancer therapy has seen substantial advances over the last few decades (1,2),predicting breast cancer risk in the apparently normal breast is still problematic (3–9). Although

Correspondence should be addressed to T.Y. ([email protected]).•Aejaz Nasir is a joining first author.

NIH Public AccessAuthor ManuscriptBreast Cancer Res Treat. Author manuscript; available in PMC 2011 January 1.

Published in final edited form as:Breast Cancer Res Treat. 2010 January ; 119(2): 335–346. doi:10.1007/s10549-009-0344-y.

NIH

-PA Author Manuscript

NIH


NIH


a few pre-malignant histologic risk factors have been identified (atypical ductal hyperplasia(ADH), lobular carcinoma in situ, microcalcifications) (10,11), few tools exist to distinguishthe normal breast from the breast at risk for cancer (3–9). Furthermore, in patients who aretreated for invasive breast cancer, the risk of local recurrence remains in spite of histologicallynegative margins. Wapnir et al (12) observed 10-year cumulative local recurrence rates rangingfrom 4.8% to 10.1% across five National Surgical Adjuvant Breast and Bowel Project(NSABP) trials involving 2,669 node-positive patients treated between 1984 and 1994, and10-year local recurrence rates of 3.5% to 6.5% were observed in node-negative patientsreceiving systemic treatment in NSABP trials (13) during the same time period.

Recent developments of gene signatures for breast cancer have been reported to benefit breastcancer prognosis (14–24). Despite these efforts and those of mammographic screening, it isstill difficult to detect risk for malignant conversion of normal breast tissue (25). Several linesof evidence suggest that histologically-normal breast tissue may, in fact, harbor pre-malignantmolecular alterations in normal breast tissue adjacent to cancer at molecular level (3–7) (4,7–9). In this study, we developed an innovative approach to identify histologically-normal, butmolecularly-abnormal “IDC-like” tissue for malignant degeneration. We postulated that ahistologically-normal tissue with “tumor-like” gene expression pattern might harborsubstantial risk for future cancer development. Genes associated with these high-risk tissueswere referred to as “malignancy-risk genes”.

The goal of our study was to establish a malignancy-risk gene expression signature inhistologically-normal breast tissues obtained from patients with ipsilateral invasive breastcancer. We have developed a gene signature to assess cancer risk by first identifying a signaturefor invasive ductal carcinoma (IDC), and by then refining it using IDC-like normal tissues. Aset of 143 histologically-normal breast tissues and 42 IDC tissues, derived from 90 patientswho underwent mastectomy for ipsilateral breast carcinoma, were assessed for global geneexpression differences using microarray analysis. A signature portending tissues at risk offuture malignancy was developed from this analysis of histologically-normal breast tissues. Itsclinical association with cancer risk was first confirmed with RTPCR and then evaluated usingtwo independent external datasets.

Materials and MethodsTissues and their associated clinicopathological data

Tissues were collected in accordance with the protocols approved by the Institutional ReviewBoard of the University of South Florida, and stored in the tissue bank of Moffitt Cancer Center.The tissues were embedded in Tissue-Tek® O.C.T., 5-µm sections cut and mounted onMercedes Platinum StarFrost™ Adhesive slides. The slides were stained using a standard H&Eprotocol, and tissue boundaries marked. Using the marked slide as a “map”, tissues weremicrodissected. Adipose tissues were trimmed away. Both histologically-normal breast tissuesand IDCs were derived from 90 patients that underwent mastectomy for various stages of breastcarcinoma and were collected and frozen in liquid nitrogen. Clinico-pathological data from thepatients used in the study, including the tumor ER, PR and Her2/Neu status and tumor grade,are shown in Table 1. When possible, each mastectomy specimen was prosected to yield anIDC and up to five sequentially-derived, adjacent normal tissue samples in the ipsilateral breastor from the four quadrants of the contralateral breast. As a result, we collected 42 IDCs and143 normal breast tissues from the 90 patients for microarray analysis. Due to RNA qualityissue in some IDC and normal tissues, we did not have a complete set of IDC and normal tissuesfor some patients. There were 11 patients (a total of 34 tissues) with at least one normal andone IDC tissue, 19 patients (a total of 28 tissues) with IDC tissue only, and 60 patients (a totalof 123 tissues) with normal tissue only. Supplementary Table 1 lists number of normal andIDC tissues and their geographical locations relative to the incident tumor.

Chen et al. Page 2

Breast Cancer Res Treat. Author manuscript; available in PMC 2011 January 1.

NIH


NIH


NIH


HistologyBased on the histopathologic review by one breast pathologist (AN), all of the 143histologically normal breast tissues were confirmed to be free of atypical ductal hyperplasia(ADH) and in-situ or invasive breast carcinoma. The 42 IDC tissues were also confirmed bythe histopathologic review by the same pathologist, based on the modified Bloom andRichardson grading scheme (26).

RNA ExtractionTotal RNA was extracted from the breast tissues using the Trizol method. Briefly, tissues werepulverized in liquid nitrogen, resuspended in 5 ml of lysis buffer, incubated for 3 min. at roomtemperature, and centrifuged at 11,500 g for 15 minutes at 4°. The aqueous phase was removedand put into another tube with 2.5 ml of isopropanol, mixed well and set at −20° C for 20minutes. The amount of RNA was quantitated by measuring A260. Microarray analysis wasperformed using the Affymetrix U133Plus 2.0 GeneChips (54,675 probe sets). Expressionvalues were calculated using the robust multi-array average (RMA) algorithm (27) (Data is inthe GEO repository: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE10780).

RT-PCR ValidationValidation of 30 selected malignancy risk signature genes (of 117 available) (SupplementaryTable 2) was done using the TaqMan Low Density Arrays (Applied Biosystems, Foster City,CA, USA). Due to limitation of sample availability, 5 “IDC-like” normal tissues, 8 IDCs, and8 normal tissues were used for validation. Single stranded cDNA was synthesized from 1 ugof total RNA using random primers in a 20 uL reaction volume using Applied Biosystem’sHigh Capacity cDNA Reverse Transcription kit. The 20 uL reactions were incubated in athermal cycler for 10 min at 25°C, 120 min at 37°C, 5 sec at 85°C and then held at 4°C. Real-time PCR was carried out using sequence specific primers/probes on the Applied Biosystems7900 HT Real-Time PCR system. cDNA was diluted 2.5-fold; 5.0 uL of diluted cDNA wasmixed with 45 uL of nuclease-free water and was added to 50 uL of TaqMan Universal PCRMaster Mix (Applied Biosystems). The 100 uL total reaction mixture was loaded in thecorresponding ports of a TaqMan Low Density Array (TLDA) card. Each TLDA card consistedof 3 replicates (4 samples per card). Expression value (ΔCt) was calculated by first averagingreplicates for each gene and then normalized (subtraction) by an endogenous control gene(18S). Since a lower value of ΔCt indicates a higher expression, a -ΔCt was used to correlatewith microarray gene expression.

Signature Generation/Statistical Methods—Statistical analysis included a series ofsteps to develop and validate the malignancy-risk gene signature (Figure 1):

1. Identification of IDC gene signature: In this first step, a set of 1038 genes (1,554 probesets) was identified that distinguished the IDCs (n = 42) from the histologically-normal tissues(n = 143). The IDC gene set was identified by treating IDC and normal tissues as twoindependent groups (although some were derived from the same patients) and using StatisticalAnalysis of Microarray (28) at 1% false discovery rate (FDR) with a fold change >2 (Figure1). The study aimed to collect multiple normal and IDC tissues from the same subjects, butdue the heterogeneous nature of the sample set, some patients had only normal tissues sampledwhile others samples were limited to IDC tissues only. This nature of unbalanced data madeit difficult to adjust for subject variation. Instead, we aggregated data into normal and IDC twogroups for comparison. To ensure homogeneity for data aggregation, we checked whetheroverall gene expression from the normal tissues in patients with normal tissues available onlywas similar to the normal tissues in patients with both normal and IDC tissues available. Weused Kmeans approach to classify all the normal tissues into two groups based on gene

Chen et al. Page 3


NIH


NIH


NIH


http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE10780

expression data. Fisher exact test did not show the two types of normal tissues were statisticallydifferent (p=0.53). We found similar results for the IDC tissues (p=0.99). These resultssuggested homogeneity for the two types of normal tissues (also for the IDC tissues).

2. Identification of “IDC-like” normal tissues: In this step, we used the IDC gene signatureto identify 11 histologically normal breast tissues that had acquired the molecular fingerprintof IDC. The method first ranked all the normal tissues for each IDC tumor gene. (e.g., A normaltissue A is ranked as the top 1% (percentile rank=100%) for tumor gene X1, top 10% (percentilerank=90%) for tumor gene X2, top 20% for tumor gene X3, and so on). As a result, for the up-regulated IDC tumor genes (e.g., k1 genes), we will have a set (k1) of the tissue percentile ranksfor each tissue. If a normal tissue displayed at least half (>k1/2) of the percentile ranks over80% (i.e., the median percentile rank>0.8), we considered it as “IDC-like” normal tissue.Similarly, a normal tissue was also considered as an IDC-like tissue if a normal tissue had themedian of the percentile ranks below 20% for down-regulated IDC tumor genes. A graphicalpresentation of the method is included in the Supplementary Figure 1. A simulation wasconducted and showed its effectiveness to identify IDC-like tissues (Supplementary Figure 2).We also compared to other approaches and results did not show these alternative approacheswere as effective as our rank approach (Supplementary Figure 3).

3. Derive malignancy-risk gene score: Once the IDC-like normal tissues were identified, wethen formed a common set of genes, “malignancy-risk signature genes”, whose expressionpercentile rank was greater than 80% (or less than 20%) in most IDC-like normal tissues. Usingthe principal components analysis (PCA) method, we derived a “risk score” (malignancy-riskscore) to represent an overall gene expression level for the malignancy-risk gene signature.First, we performed principal components analysis to reduce data dimension into a small setof uncorrelated principal components. This set of principal components was generated basedon its ability to account for variation. We used the first principal component (1st PCA), as itaccounts for the largest variability in the data, as a malignancy risk score to represent theoverall expression level for the signature. That is, malignancy risk score=∑wixi, a weightedaverage expression among the malignancy-risk genes, where xi represents gene i expression

level, wi is the corresponding weight with , and the wi values maximize the varianceof ∑wixi . While other PCAs (e.g., the second and third principle components) may alsoassociate with cancer risk, our experiences indicates that the 1st PCA often corresponds mosteffectively to cancer risk-related information for this study (see RT-PCR and DCIS validationin the Results section).

4. Cross-validation: Leave-one-out cross validation (LOOCV) was performed to evaluaterobustness of the IDC and malignancy-risk gene signatures. This was done by excluding onesample at a time and repeating steps 1–3 to see how many were correctly identified (IDC genes,IDC-like normal tissues, and malignancy-risk genes).

5. Pathway analysis: Pathway analysis was done using MetaCore ™ by GeneGo for steps 1and 3 to identify biological functions associated with IDC genes and the malignancy-risk genes.We compared pathways of the two gene sets to reveal difference of biological processesbetween the IDC genes and the malignancy-risk genes.

6. RT-PCR validation: Pearson correlation was used to evaluate association of the malignancyrisk score between microarray and RT-PCR platforms. The malignancy-risk score wascalculated using the 30 selected malignancy-risk signature genes (see Statistical Methods) formicroarray and RT-PCR, respectively. Correlation analysis was also performed for eachindividual malignancy-risk gene. Analysis of variance was used to test the differences amongthe three groups (normal, IDC-like normal, and IDC) with the Tukey method (29) to adjust for

Chen et al. Page 4


NIH


NIH


NIH


p value for pair-wise comparison. We also used support vector machine (SVM) to build aclassifier from the microarray platform to evaluate the prediction performance on the RT-PCRplatform.

7. Evaluation of cancer risk and cancer progression: We assessed the cancer risk potentialand the cancer progression of the malignancy-risk score on two independent data sets. Becauseeach data set had a different set of available genes, we used whatever genes were in commonwith the malignancy risk gene signature to evaluate each data set (essentially a subset of theoriginal malignancy-risk gene signature). The SVM was used to evaluate predictionperformance. In addition, for ordinal clinical variable (e.g., from normal, ductal carcinoma insitu (DCIS), to IDC), the malignancy-risk score was used to correlate with cancer severityusing Pearson correlation to evaluate the trend of the malignancy-risk gene signature withcancer progression.

ResultsIDC gene signature

An IDC gene signature (1,554 probe sets: 1038 unique genes) was developed from a set of 42IDC and 143 normal breast tissues using Statistical Analysis of Microarray (28). We found theIDC gene set remained robust in the leave-one-out cross-validation. Pathway analysis revealedtwo predominant cellular processes: cell adhesion and cell cycle/proliferation (SupplementaryTable 3).

IDC-like normal breast tissuesWe ranked the 143 normal breast tissues based on the IDC gene signature and identified 11“IDC-like” normal breast tissues, whose gene expression profiles more closely approximatedthat of the IDC samples rather than the rest of the 132 normal breast tissues (i.e., these 11 IDC-like normal tissues were molecularly- abnormal, but histologically-normal) (SupplementaryFigure 4).

Histology of IDC-like normal tissuesSupplementary Table 4 summarizes the normal histological findings of the 11 IDC-like normalbreast tissues used in this study. All of these specimens consisted of completely unremarkable,benign breast tissues and were free from in-situ or invasive carcinoma as well as atypical ductalhyperplasia (Figure 2). Fisher exact test showed no significant association of patients harboringIDC-like normal tissues with ER/PR/Her2/grade (Supplementary Table 5).

Malignancy-risk gene signature and risk scoreA malignancy-risk gene signature was developed by forming a “common set” of genes whoseexpression varied (up or down) at high levels in the 11 IDC-like normal tissues (see StatisticalMethods). The malignancy-risk genes consisted of 109 up-regulated probe sets (96 uniquegenes) and 31 down-regulated probe sets (21 unique genes). Table 2 provides a selected subsetof malignancy-risk genes; the entire list is presented in Supplementary Table 6. Moreover, byutilizing principal component analysis, a malignancy-risk score was derived to represent anoverall gene expression level for the malignancy-risk signature (see Statistical Methods).

Cross-validationAnalysis of the malignancy risk score by LOOCV yielded a high degree of consistency; mostIDC genes (>98%), IDC-like normal tissues (>90%), and malignancy-risk genes (>90%) wereidentified correctly at each leave-one-out iteration (Supplementary Figure 5). Moreover, ateach iteration, we calculated a predicted malignancy risk score for the sample being excluded.

Chen et al. Page 5


NIH


NIH


NIH


Correlation analysis showed a strong relationship between the predicted risk score and thedisease status (i.e., by ranking normal, IDC-like normal, and IDC from 0 to 2; Pearsoncorrelation=0.89 and Spearman correlation=0.74 with p <0.0001).

Pathway analysis of malignancy-risk genesIn contrast to the IDC gene signature, pathway analysis of the malignancy-risk gene set showeda remarkable over-expression of proliferative function genes, instead of a mixture ofproliferation and adhesion genes seen with IDC. There were 11 cell cycle related pathwaysrepresented in the malignancy-risk signature (p value <0.01, Supplementary Table 7). Sincethe malignancy-risk gene signature was derived from the IDC gene signature, the differencein functional classes of genes would not have been expected in the absence of a selectionbias. The majority of the malignancy-risk genes were classified to be primarily associated withDNA replication and mitosis, two hallmark events associated with proliferation(Supplementary Table 8). This observation may indicate the importance of these features inearly stages of tumorigenesis (30).

Weak correlation of malignancy risk score with ER, PR, and Her2Since ER, PR, and Her2 are key markers in cancer development, we examined their correlationwith the malignancy risk score. Results showed only a weak correlation for ER and PR (r=−0.2~0.3) and a moderate correlation with Her2 (r=0.37~0.47 by spearman correlation andr=0.43~0.63 by Pearson correlation), suggesting relative independence of the risk score fromthese biomarkers (Supplementary Figure 6).

Higher malignancy risk score of IDC-like normal tissuesWe identified 11 IDC-like normal tissues from 10 patients. There were another 12 normaltissues collected from the same 10 patients. These 12 normal tissues were molecularly andhistologically normal and labeled as matched normal tissues to reflect they were derived fromthe same subject. The other normal tissues (n=120) from subjects without IDC-like normaltissues (i.e., not from the 10 subjects) were also molecularly and histologically normal andlabeled as unmatched normal tissues for distinction. Interestingly, we found the malignancyrisk score was higher in the IDC-like normal tissues and the matched normal tissues than inthe unmatched normal tissues. Difference of the risk score was statistically significant for (a)IDC-like normal tissues versus the matched normal tissues (adjusted p value<0.0001 using theTukey method) and (b) matched versus unmatched normal tissues (adjusted p value=0.0054).An increasing trend of the malignancy risk score was also seen from the unmatched normaltissues, the matched normal tissue, to the IDC-like normal tissues at the pooled data level(Pearson correlation=0.63 with p<0.0001; Figure 3). Moreover, among the 10 patients withIDC-like normal tissues, analysis results showed a higher malignancy risk score in the IDC-like normal tissues than in the matched normal tissues at subject level (p=0.01 using the randomeffect model; Figure 3). Since the malignancy risk score was derived without knowing subjectinformation, a trend of the risk score decreasing from the IDC-like normal tissues, to thematched normal tissue, to the unmatched normal tissues would not be expected.

RT-PCR validation of malignancy-risk genesExpression of the 30 selected malignancy risk signature genes identified by microarrayprofiling was successfully validated by RT-PCR. There were 27 genes showing a strongPearson correlation >0.7 (correlation>0.9: 12 genes, 0.8–0.9: 13, and 0.7–0.8: 2; the p valueswere <0.0001) (Supplementary Figure 7). The composite malignancy risk score (based onmicroarray data from 30 genes) also demonstrated a very high correlation (0.95) with RT-PCRresults. The risk score for the IDC-like normal tissues fell in the middle between the IDC andnormal samples (Figure 4). In addition, we used support vector machine (SVM) to build a

Chen et al. Page 6


NIH


NIH


NIH


classifier for the 30 genes from microarray (accuracy rate=86% using LOOCV) and predictedon the RT-PCR expression with 90% accuracy (Supplementary Figure 7). In comparing themalignancy risk score generated by various PCAs, the use of the 1st PCA as the risk scoreshowed a very high correlation (r=0.95) between microarray and RT-PCR and an increasingtrend of the risk score from normal to IDC samples. The other PCAs had a weak correlation(r<0.5) and did not yield their association with cancer progression.

Validation of Moffitt ductal carcinoma in situ (DCIS) samples for cancer progressionA set of 23 DCIS samples (from 11 patients: 8 from the 90 patients and 3 new patients) werecollected from the Moffitt Cancer Center to evaluate the cancer progression feature of themalignancy-risk gene signature. We compared the malignancy risk score among the fourgroups: normal breast, IDC-like normal, DCIS, and IDC. Data showed an increasing risk scorepattern with progression from normal, to IDC-like normal, to DCIS, and to IDC. Pearson andSpearman correlation was 0.87 and 0.8, respectively, with a significant p value<0.0001 (byranking the disease status from 0 to 3 for normal breast to IDC). Moreover, the malignancy-risk score of DCIS was lower than IDC, but higher than normal tissue (p =0.0005) within eachpatient (Figure 5). In addition, we evaluated the prediction performance on the DCIS samples.A SVM classifier was built with an accuracy rate of 92% by 10-fold cross validation. Theclassifier predicted most of the 23 DCIS samples into the IDC category (18/23) and two samplesfavoring to the “IDC-like normal” group (Supplementary Figure 8). We also compared themalignancy risk score generated by PCA1 (1st PCA) versus other PCAs (PCA2 and PCA3).In contrast to PCA1 showing a cancer progression pattern, PCA 2 and PCA3 did notdemonstrate a cancer progression from non-IDC like normal to IDC (Supplementary Figure8).

Evaluation of cancer risk in Poola et al’s ADH study (31)This study was selected in order to assess the potential of the malignancy-risk score to predictthe risk of future cancer development in the breast associated with ADH. The study collected4 ADH tissues from patients without breast cancer development (labeled as ADH-N) and 4ADH samples with cancer developed (labeled as ADH-C). We used 102 probe sets from theirplatform (in common with the malignancy-risk gene signature) to calculate the malignancyrisk score by the 1st PCA and the 2nd PCA, respectively. SVM was then used to classify the 8patients based on the malignancy-risk score. Data analysis showed the use of the 1st PCA asthe malignancy risk score yielding a higher risk score in the ADH-C group than in the ADH-N group (Figure 6). The SVM correctly classified 7 out 8 patients (87.5%) although it was notstatistically significant (p= 0.14 based on the fisher exact test) due to a very limited samplesize (n=4 per group). Notably, three out the four ADH-C patients had a risk score above 5 witha higher cancer-risk probability, in contrast to most ADH-N patients with negative scores anda lower cancer-risk probability. As the 2nd PCA was incorporated with the 1st PCA in themodel, SVM correctly classified all the 8 patients (Supplementary Figure 9), suggesting themalignancy-risk gene signature was able to differentiate ADH patients between with andwithout cancer development, and indicating its ability to assessing cancer risk. We alsocalculated area under curve (AUC) for response operating characteristic curve. The resultswere similar to the ones by SVM: AUC was higher by the use of the first PCAs (AUC=1) thanby the 1st PCA (AUC=0.875) (Supplementary Figure 9).

DiscussionIdentification of normal tissue at risk for malignant conversion has great potential applicationin clinical practice, in both evaluating the malignancy risk measured by routine breast biopsiesas well as the risk of local recurrence following lumpectomy. Detecting these high-risk normal,appearing tissues, however, remains a challenging task. In this study, we utilized the IDC data

Chen et al. Page 7


NIH


NIH


NIH


information to develop a new concept of "IDC-like normal tissue". The "IDC-like normaltissue" could be histologically normal tissue with a molecular proclivity towards IDC. Basedon this hypothesis, we identified 11 “IDC-like” normal tissues (out of 143 normal breast tissues)and developed the malignancy-risk gene signature and risk score.

A careful re-examination of all the IDC-like normal tissues showed that they werehistologically-normal, with no evidence of in-situ or invasive carcinoma of the breast, and noatypia (Figure 2 and Supplementary Table 4). However, these IDC-like normal tissues showedgene expression profiles resembling invasive carcinomas, indicating that these tissues hadalready acquired the molecular fingerprint of cancer and, therefore, may be at increased riskfor subsequent cancer development. Moreover, from these IDC-like normal tissues, wedeveloped a “malignancy-risk” gene signature that may serve as a marker of subsequent riskof breast cancer development. The malignancy-risk gene signature was internally validated byRT-PCR and leave-one-out cross validation as well as by two additional datasets. Furtheranalysis of external datasets also demonstrated its clinical relevance to cancer-risk and cancerprogression. While this gene signature requires further clinical validation, this is an intriguingfinding with substantive clinical implications. Several studies have suggested that cell cycle/proliferation are key hallmarks of existing cancer (22,34–36). This is the first study, however,to suggest the proliferative program of gene expression may be the earliest detectable event innormal breast tissues at risk for developing breast cancer. Moreover, this is the largestmolecular analysis of histologically benign breast tissues. A recently reported study of 14normal breast tissues from breast cancer cases identified genes differentially expressed in thesetissues versus normal breast reduction mammoplasties, but did not decipher a predominantlyproliferative gene function (37).

The large preponderance of proliferative genes in the malignancy-risk gene set was notexpected. By comparison, IDC associated genes were biased towards both proliferative andadhesive gene sets. These findings suggest a temporal relationship between proliferative andadhesive gene expression programs, with the former being precursors to histological alterationsand responsible for malignancy risk. Interestingly, there was also no statistical association ofthe IDC-like normal tissues with ER/PR, Her2/neu, and grade suggesting the malignancy risksignature may be not be dependent on these factors. The lack of association of the IDC-likenormal tissues with the triple negative (ER/PR/Her2Neu) phenotype also suggests no link toBRCA1 and BRCA2.

Evaluation on two external independent datasets demonstrated the clinical relevance of themalignancy-risk gene signature to cancer risk. While further validation of the malignancy-risksignature is warranted, the signature has promise for impacting clinical decisions. Theseinclude altering strategies for follow-up of histologically-normal, but molecularly abnormalbreast biopsies, determining which patients might benefit from radiotherapy followinglumpectomy, or determining which patients might benefit from mastectomy due to multifocaldisease risk.

Supplementary MaterialRefer to Web version on PubMed Central for supplementary material.

AcknowledgmentsThis Research is funded by National Cancer Institute grant R01CA112215 (PI: TJY) and National Cancer InstituteGrant R01 CA 098522(PI: TJY)

Chen et al. Page 8


NIH


NIH


NIH


References1. Berry DA, Cronin KA, Plevritis SK, Fryback DG, Clarke L, Zelen M, et al. Effect of screening and

adjuvant therapy on mortality from breast cancer. N Engl J Med 2005;353(17):1784–1792. [PubMed:16251534]

2. Giordano SH, Buzdar AU, Smith TL, Kau SW, Yang Y, Hortobagyi GN. Is breast cancer survivalimproving? Cancer 2004;100(1):44–52. [PubMed: 14692023]

3. Lewis CM, Cler LR, Bu DW, Zochbauer-Muller S, Milchgrub S, Naftalis EZ, et al. Promoterhypermethylation in benign breast epithelium in relation to predicted breast cancer risk. ClinicalCancer Research 2005;11(1):166–172. [PubMed: 15671542]

4. Deng GR, Lu Y, Zlotnikov G, Thor AD, Smith HS. Loss of heterozygosity in normal tissue adjacentto breast carcinomas. Science 1996;274(5295):2057–2059. [PubMed: 8953032]

5. Fredriksson I, Liljegren G, Palm-Sjovall M, Arnesson LG, Emdin SO, Fornander T, et al. Risk factorsfor local recurrence after breast-conserving surgery. British Journal of Surgery 2003;90(9):1093–1102.[PubMed: 12945077]

6. Ellsworth DL, Ellsworth RE, Love B, Deyarmin B, Lubert SM, Mittal V, et al. Outer breast quadrantsdemonstrate increased levels of genomic instability. Annals of Surgical Oncology 2004;11(9):861–868. [PubMed: 15313734]

7. Botti C, Pescatore B, Mottolese M, Sciarretta F, Greco C, Di Filippo F, et al. Incidence of chromosomes1 and 17 aneusomy in breast cancer and adjacent tissue: An interphase cytogenetic study. Journal ofthe American College of Surgeons 2000;190(5):530–539. [PubMed: 10801019]

8. Larson PS, de las Morenas A, Bennett SR, Cupples LA, Rosenberg CL. Loss of heterozygosity or alleleimbalance in histologically normal breast epithelium is distinct from loss of heterozygosity or alleleimbalance in co-existing carcinomas. American Journal of Pathology 2002;161(1):283–290.[PubMed: 12107113]

9. Li Z, Moore DH, Meng ZH, Ljung BM, Gray JW, Dairkee SH. Increased risk of local recurrence isassociated with allelic loss in normal lobules of breast cancer patients. Cancer Research 2002;62(4):1000–1003. [PubMed: 11861372]

10. Schnitt SJ, Morrow M. Lobular carcinoma in situ: current concepts and controversies. Semin DiagnPathol 1999;16(3):209–223. [PubMed: 10490198]

11. Fitzgibbons PL, Henson DE, Hutter RV. Benign breast changes and the risk for subsequent breastcancer: an update of the 1985 consensus statement. Cancer Committee of the College of AmericanPathologists. Arch Pathol Lab Med 1998;122(12):1053–1055. [PubMed: 9870852]

12. Wapnir IL, Anderson SJ, Mamounas EP, Geyer CE Jr, Jeong JH, Tan-Chiu E, et al. Prognosis afteripsilateral breast tumor recurrence and locoregional recurrences in five National Surgical AdjuvantBreast and Bowel Project node-positive adjuvant breast cancer trials. J Clin Oncol 2006;24(13):2028–2037. [PubMed: 16648502]

13. Wapnir SA, Mamounas E, Geyer C, Hyeon-Jeong J, Costantino J, Fisher B, Wolmark N. Survivalafter IBTR in NSABP Node Negative Protocols B-13, B-14, B-19, B-20 and B-23. Journal of ClinicalOncology 2005;23(No 16S):517.

14. Carter SL, Eklund AC, Kohane IS, Harris LN, Szallasi Z. A signature of chromosomal instabilityinferred from gene expression profiles predicts clinical outcome in multiple human cancers. NatureGenetics 2006;38(9):1043–1048. [PubMed: 16921376]

15. Chang HY, Nuyten DS, Sneddon JB, Hastie T, Tibshirani R, Sorlie T, et al. Robustness, scalability,and integration of a wound-response gene expression signature in predicting breast cancer survival.Proc Natl Acad Sci U S A 2005;102(10):3738–3743. [PubMed: 15701700]

16. Finak G, Bertos N, Pepin F, Sadekova S, Souleimanova M, Zhao H, et al. Stromal gene expressionpredicts clinical outcome in breast cancer. Nat Med 2008;14(5):518–527. [PubMed: 18438415]

17. Glas AM, Floore A, Delahaye LJ, Witteveen AT, Pover RC, Bakx N, et al. Converting a breast cancermicroarray signature into a high-throughput diagnostic test. BMC Genomics 2006;7:278. [PubMed:17074082]

18. Huang E, Cheng SH, Dressman H, Pittman J, Tsou MH, Horng CF, et al. Gene expression predictorsof breast cancer outcomes. Lancet 2003;361(9369):1590–1596. [PubMed: 12747878]

Chen et al. Page 9


NIH


NIH


NIH


19. Ma XJ, Salunga R, Tuggle JT, Gaudet J, Enright E, McQuary P, et al. Gene expression profiles ofhuman breast cancer progression. Proceedings of the National Academy of Sciences of the UnitedStates of America 2003;100(10):5974–5979. [PubMed: 12714683]

20. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. A multigene assay to predict recurrence oftamoxifen-treated, node-negative breast cancer. N Engl J Med 2004;351(27):2817–2826. [PubMed:15591335]

21. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, et al. Molecular portraits ofhuman breast tumours. Nature 2000;406(6797):747–752. [PubMed: 10963602]

22. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, et al. Gene expression patterns ofbreast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U SA 2001;98(19):10869–10874. [PubMed: 11553815]

23. van de Vijver MJ, He YD, van 't Veer LJ, Dai H, Hart AAM, Voskuil DW, et al. A gene-expressionsignature as a predictor of survival in breast cancer. New England Journal of Medicine 2002;347(25):1999–2009. [PubMed: 12490681]

24. Wang Y, Klijn JG, Zhang Y, Sieuwerts AM, Look MP, Yang F, et al. Gene-expression profiles topredict distant metastasis of lymph-node-negative primary breast cancer. Lancet 2005;365(9460):671–679. [PubMed: 15721472]

25. Shah VI, Raju U, Chitale D, Deshpande V, Gregory N, Strand V. False-negative core needle biopsiesof the breast - An analysis of clinical, radiologic, and pathologic findings in 27 consecutive cases ofmissed breast cancer. Cancer 2003;97(8):1824–1831. [PubMed: 12673707]

26. Robbins P, Pinder S, Deklerk N, Dawkins H, Harvey J, Sterrett G, et al. Histological grading of breastcarcinomas - a study of interobserver agreement. Human Pathology 1995;26(8):873–879. [PubMed:7635449]

27. Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. Summaries of AffymetrixGeneChip probe level data. Nucleic Acids Res 2003:e15. [PubMed: 12582260]

28. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiationresponse. Proceedings of the National Academy of Sciences of the United States of America 2001;98(9):5116–5121. [PubMed: 11309499]

29. Miller, RG. Simultaneous Statistical Inference. Springer; 1981.30. Whitfield ML, Sherlock G, Saldanha AJ, Murray JI, Ball CA, Alexander KE, et al. Identification of

genes periodically expressed in the human cell cycle and their expression in tumors. MolecularBiology of the Cell 2002;13(6):1977–2000. [PubMed: 12058064]

31. Poola I, DeWitty RL, Marshalleck JJ, Bhatnagar R, Abraham J, Leffall LD. Identification of MMP-1as a putative breast cancer predictive marker by global gene expression analysis. Nature Medicine2005;11(5):481–483.

32. Turashvili G, Bouchal J, Baumforth K, Wei W, Dziechciarkova M, Ehrmann J, et al. Novel markersfor differentiation of lobular and ductal invasive breast carcinomas by laser microdissection andmicroarray analysis. Bmc Cancer 2007:7. [PubMed: 17222331]

33. Chanrion M, Negre V, Fontaine H, Salvetat N, Bibeau F, Mac Grogan G, et al. A gene expressionsignature that can predict the recurrence of tamoxifen-treated primary breast cancer. Clin Cancer Res2008;14(6):1744–1752. [PubMed: 18347175]

34. Rosenwald A, Wright G, Wiestner A, Chan WC, Connors JM, Campo E, et al. The proliferation geneexpression signature is a quantitative integrator of oncogenic events that predicts survival in mantlecell lymphoma. Cancer Cell 2003;3(2):185–197. [PubMed: 12620412]

35. Whitfield ML, George LK, Grant GD, Perou CM. Common markers of proliferation. Nat Rev Cancer2006;6(2):99–106. [PubMed: 16491069]

36. Chung CH, Bernard PS, Perou CM. Molecular portraits and the family tree of cancer. Nat Genet 2002;(32 Suppl):533–540. [PubMed: 12454650]

37. Tripathi A, King C, de la Morenas A, Perry VK, Burke B, Antoine GA, et al. Gene expressionabnormalities in histologically normal breast epithelium of breast cancer patients. Int J Cancer2008;122(7):1557–1566. [PubMed: 18058819]

Chen et al. Page 10


NIH


NIH


NIH


Figure 1.Flow chart to developing the malignancy-risk gene signature.

Chen et al. Page 11


NIH


NIH


NIH


Figure 2. Histologic images of representative frozen breast tissues (original magnification X 200)(a) Invasive ductal carcinoma (IDC) showing sheets of tumor cells and stromal strands, (b)Normal breast lobule in a frozen breast tissue specimen that was collected at 1 cm from thetumor (IDC) shown in Figure a. This specimen was designated as ‘IDC-like normal’ based onits molecular profile, (c) Normal breast lobule in a frozen breast tissue specimen that wascollected at 2 cm from the tumor (IDC) shown in Figure a.

Chen et al. Page 12


NIH


NIH


NIH


Chen et al. Page 13


NIH


NIH


NIH


Figure 3. Comparison of malignancy-risk score between IDC-like normal tissues, their matchednormal tissues, and unmatched normal tissuesMalignancy risk score was compared among the three groups in both subject level and pooleddata level (ignore subject information). (a) Subject level: A higher malignancy risk score wasseen in the IDC-like normal tissues than in the matched normal tissues for most subjects (p=0.01using the random effect model); (b) Pooled data level: An increasing trend of the malignancyrisk score was seen from the unmatched normal tissues, the matched normal tissue, to the IDC-like normal tissues (Pearson correlation=0.63 with p<0.0001); (c) Pair-wise comparison of themalignancy risk score among the unmatched normal tissues, the matched normal tissue, andthe IDC-like normal tissues using the Tukey method to control for the type I error.

Chen et al. Page 14


NIH


NIH


NIH


Figure 4. RT-PCR validationPearson correlation was used to evaluate association of the malignancy risk score betweenmicroarray and RT-PCR. The malignancy-risk score was calculated using the 30 selectedmalignancy risk genes (Supplementary table 2) for microarray and RT-PCR, respectively.

Chen et al. Page 15


NIH


NIH


NIH


Figure 5. Validation of Moffitt ductal carcinoma in situ (DCIS) samples for cancer progressionThe malignancy-risk score was generated by the 1st PCA and showed an increasing trend fromnon-IDC-like normal, IDC-like normal, to IDC. Additional analysis results were inSupplementary Figure 7.

Chen et al. Page 16


NIH


NIH


NIH


Figure 6. Evaluation for cancer risk in Poola et al’s ADH studyThe use of the 1st PCA as the malignancy risk score showed that the ADH-C group had a higherrisk score than the ADH-N group (Figure 6A) with AUC=0.875 (Figure 6B) . Additionalanalysis results were in Supplementary Figure 8.

Chen et al. Page 17


NIH


NIH


NIH


Table 1

Pathological data of the patients used in the study, including ER, PR, Her2, and grade.

ER/PR/Her2 status

ER R P u Her2/ne

Negative 25 38 43

Positive 55 42 12

other* 10 10 35

Total cases 90 90 90

Grade frequency

Well differentiated 6

Moderately differentiated 27

Poorly differentiated 30

Undifferentiated/anaplastic 10

No grade 17

Total cases 90*Results not available

Chen et al. Page 18


NIH


NIH


NIH


Tabl

e 2

A su

bset

of m

alig

nanc

y-ris

k ge

nes a

ssoc

iate

d w

ith D

NA

repl

icat

ion,

mito

sis,

and

canc

er ri

sk * .

Affy

pro

be se

t id

Gen

e Sy

mbo

lFo

ld c

hang

eFD

RR

egul

atio

nD

NA

rep

licat

ion

Mito

sis

Pool

a et

al

Gen

e T

itle

2226

08_s

_at

AN

LN4.

01<0

.01

Up-

Reg

ulat

edY

anill

in, a

ctin

bin

ding

pro

tein

(scr

aps h

omol

og,

Dro

soph

ila)

2020

95_s

_at

BIR

C5

2.95

<0.0

1U

p-R

egul

ated

Yba

culo

vira

l IA

P re

peat

- con

tain

ing

5 (s

urvi

vin)

2096

42_a

tB

UB

12.

71<0

.01

Up-

Reg

ulat

edY

BU

B1

budd

ing

unin

hibi

ted

by b

enzi

mid

azol

es 1

hom

olog

(yea

st)

2037

55_a

tB

UB

1B3.

05<0

.01

Up-

Reg

ulat

edY

BU

B1

budd

ing

unin

hibi

ted

by b

enzi

mid

azol

es 1

hom

olog

beta

(yea

st)

2147

10_s

_at

CC

NB

14.

03<0

.01

Up-

Reg

ulat

edY

cycl

in B

1

2027

05_a

tC

CN

B2

2.35

<0.0

1U

p-R

egul

ated

Ycy

clin

B2

2050

34_a

tC

CN

E23.

99<0

.01

Up-

Reg

ulat

edY

cycl

in E

2

2032

13_a

tC

DC

25.

5<0

.01

Up-

Reg

ulat

edY

Cel

l div

isio

n cy

cle

2, G

1 to

S a

nd G

2 to

M

2032

14_x

_at

CD

C2

2.89

<0.0

1U

p-R

egul

ated

Yce

ll di

visi

on c

ycle

2, G

1 to

S a

nd G

2 to

M

2105

59_s

_at

CD

C2

4.14

<0.0

1U

p-R

egul

ated

YY

cell

divi

sion

cyc

le 2

, G1

to S

and

G2

to M

1555

758_

a_at

CD

KN

32.

85<0

.01

Up-

Reg

ulat

edcy

clin

-dep

ende

nt k

inas

e in

hibi

tor 3

(CD

K2-

asso

ciat

eddu

al sp

ecifi

city

pho

spha

tase

)

2097

14_s

_at

CD

KN

32.

97<0

.01

Up-

Reg

ulat

edY

cycl

in-d

epen

dent

kin

ase

inhi

bito

r 3 (C

DK

2-as

soci

ated

dual

spec

ifici

ty p

hosp

hata

se)

2049

62_s

_at

CEN

PA2.

71<0

.01

Up-

Reg

ulat

edce

ntro

mer

e pr

otei

n A

, 17k

Da

2078

28_s

_at

CEN

PF2.

6<0

.01

Up-

Reg

ulat

edce

ntro

mer

e pr

otei

n F,

350

/400

ka (m

itosi

n)

2185

42_a

tC

EP55

3.46

<0.0

1U

p-R

egul

ated

Ych

rom

osom

e 10

ope

n re

adin

g fr

ame

3

2182

52_a

tC

KA

P22.

72<0

.01

Up-

Reg

ulat

edY

Ycy

tosk

elet

on a

ssoc

iate

d pr

otei

n 2

2037

64_a

tD

LG7

2.84

<0.0

1U

p-R

egul

ated

YY

disc

s, la

rge

hom

olog

7 (D

roso

phila

)

2033

58_s

_at

EZH

22.

69<0

.01

Up-

Reg

ulat

edY

enha

ncer

of z

este

hom

olog

2 (D

roso

phila

)

2139

11_s

_at

H2A

FZ2.

21<0

.01

Up-

Reg

ulat

edY

H2A

his

tone

fam

ily, m

embe

r Z

2025

03_s

_at

KIA

A01

015.

89<0

.01

Up-

Reg

ulat

edK

IAA

0101

2047

09_s

_at

KIF

232.

14<0

.01

Up-

Reg

ulat

edY

kine

sin

fam

ily m

embe

r 23

2021

07_s

_at

MC

M2

2.08

<0.0

1U

p-R

egul

ated

YM

CM

2 m

inic

hrom

osom

e m

aint

enan

ce d

efic

ient

2,

mito

tin (S

. cer

evis

iae)

2048

25_a

tM

ELK

3.76

<0.0

1U

p-R

egul

ated

YY

mat

erna

l em

bryo

nic

leuc

ine

zipp

er k

inas

e

2046

41_a

tN

EK2

5.55

<0.0

1U

p-R

egul

ated

NIM

A (n

ever

in m

itosi

s gen

e a)

-rel

ated

kin

ase

2

Chen et al. Page 19


NIH


NIH


NIH


Affy

pro

be se

t id

Gen

e Sy

mbo

lFo

ld c

hang

eFD

RR

egul

atio

nD

NA

rep

licat

ion

Mito

sis

Pool

a et

al

Gen

e T

itle

2015

77_a

tN

ME1

2.15

<0.0

1U

p-R

egul

ated

non-

met

asta

tic c

ells

1, p

rote

in (N

M23

A) e

xpre

ssed

in

2180

39_a

tN

USA

P16.

41<0

.01

Up-

Reg

ulat

edY

nucl

eola

r and

spin

dle

asso

ciat

ed p

rote

in 1

2199

78_s

_at

NU

SAP1

5<0

.01

Up-

Reg

ulat

edY

Ynu

cleo

lar a

nd sp

indl

e as

soci

ated

pro

tein

1

2220

77_s

_at

RA

CG

AP1

3.36

<0.0

1U

p-R

egul

ated

Rac

GTP

ase

activ

atin

g pr

otei

n 1

2018

90_a

tR

RM

28.

07<0

.01

Up-

Reg

ulat

edY

ribon

ucle

otid

e re

duct

ase

M2

poly

pept

ide

2097

73_s

_at

RR

M2

6.73

<0.0

1U

p-R

egul

ated

YY

ribon

ucle

otid

e re

duct

ase

M2

poly

pept

ide

2092

18_a

tSQ

LE3.

25<0

.01

Up-

Reg

ulat

edsq

uale

ne e

poxi

dase

1554

408_

a_at

TK1

2.72

<0.0

1U

p-R

egul

ated

Yth

ymid

ine

kina

se 1

, sol

uble

2023

38_a

tTK

12.

86<0

.01

Up-

Reg

ulat

edY

thym

idin

e ki

nase

1, s

olub

le

2012

91_s

_at

TOP2

A7.

56<0

.01

Up-

Reg

ulat

edY

topo

isom

eras

e (D

NA

) II a

lpha

170

kDa

2012

92_a

tTO

P2A

6.03

<0.0

1U

p-R

egul

ated

Yto

pois

omer

ase

(DN

A) I

I alp

ha 1

70kD

a

2048

22_a

tTT

K3.

27<0

.01

Up-

Reg

ulat

edY

TTK

pro

tein

kin

ase

2040

26_s

_at

ZWIN

T4.

46<0

.01

Up-

Reg

ulat

edZW

10 in

tera

ctor

* “Y”

sym

bol w

as u

sed

to in

dica

te th

e as

soci

atio

n of

eac

h m

alig

nanc

y-ris

k ge

ne w

ith D

NA

repl

icat

ion,

mito

sis,

and

canc

er ri

sk.

Chen et al. Page 20


NIH


NIH


NIH


Proliferative g enes dominate malignancy-risk gene ...

Documents

Transcript of Proliferative g enes dominate malignancy-risk gene ...